Page 98 - Read Online

Mooraj et al. J Mater Inf 2023;3:4 https://dx.doi.org/10.20517/jmi.2022.41 Page 23 of 45

Table 1. Comparison of pros, cons, and capabilities of various computational methods
Computational Predicted Pros Cons Number of screened References
method Properties compositions
5
Machine learning Elastic constants High computational Requires large training 10 [48,78,79,81,82,87,
Phase formation efficiency sets 90,137-139]
Phase transformation Versatility in predictive Lack of physical
temperature features interpretability
Hardness Only gives statistical
Tensile strength understanding
Compressive strength
4
First-principles Elastic constants Low input information Computationally 10
Phase formation needed expensive
Phase transformation Provides fundamental Time-consuming
temperature understanding
Atomic scale detail
3
Molecular dynamics Elastic constants Provides fundamental Computationally 10 [114,115,125,127,141,
Phase formation understanding expensive 142]
Phase transformation Atomic scale detail Time-consuming
temperature Dynamically simulate Cannot provide
Hardness microstructure evolution macroscopic results
Tensile strength
Compressive strength
6
CALPHAD Phase formation High computational Only predicts 10
Phase transformation efficiency equilibrium conditions
temperature High accuracy No kinetic information
Easily interpretable

information about equilibrium phase formation and transformation temperatures which may not be
representative of manufacturing or application conditions. This limitation is especially important for
HEAs,c where sluggish diffusion limits the kinetics within the system, which can often lead to the formation
of metastable phases that may not be expected under equilibrium conditions.

COMBINATORIAL ADDITIVE MANUFACTURING TO EXPLORE LARGE COMPOSITIONAL
SPACE
After narrowing a target composition space using computational methods, the remaining candidate
compositions are still too numerous to reasonably explore via traditional metallurgical techniques. Thus,
high-throughput manufacturing techniques are needed to rapidly produce samples that cover the candidate
composition region. Previous studies have utilized magnetron sputtering and diffusion multiples to produce
combinatorial libraries [28,149-151] . However, as previously discussed, these techniques produce samples at
micro- or nano-scale, which may not be representative of bulk materials.

Additionally, the cooling rates experienced during magnetron sputtering are orders of magnitude greater
[35,36]
than the cooling rates in traditional manufacturing settings . Thus, there is a need for a manufacturing
technique that can produce vast compositional libraries at a bulk length scale with practically relevant
cooling rates. Laser additive manufacturing (LAM) has shown great promise towards that end. Previously
LAM has been used to produce alloys with improved properties compared to their conventionally
[36,152-159]
manufactured counterparts . Two main types of LAM are used in combinatorial studies, i.e., laser
directed energy deposition (DED), also known as laser engineered net shaping (LENS), and laser powder-
[160]
bed fusion (L-PBF) . The DED process utilizes a carrier gas that allows the powder to flow continuously
while shielding it from oxidation during deposition. A laser source simultaneously heats the material upon
[37]
contact with the printing substrate or previous layer . In the case of L-PBF, a flatbed of powder is
deposited on a substrate. A laser is then used to melt the particles in a pattern determined by design
[161]
software to form a part layer by layer .